Summaries
At the end of a debate or Ask request, Palabre can ask an agent to produce a final summary.
Summary format
Debate requests four sections: Consensus, Disagreements / uncertainties, Proposed actions, Conclusion.
Ask requests five sections: Faithful summary by agent, Comparison, Watchpoints / uncertainties, Proposed actions, Conclusion.
These are the actual English prompt labels. The first sections are lists; Conclusion is a short prose paragraph. Chat and Relay add no final summary; Chat can continue from an existing summary.
Summary agent
In debate mode, Palabre uses defaults.summaryAgent when it exists. Otherwise, it uses agent B.
In Ask mode, Palabre uses defaults.askSummaryAgent when it exists, then defaults.summaryAgent, then the last Ask agent.
To choose the summary agent for a command:
palabre codex-claude "Critique this plan" --summary-agent claude
To configure a summary agent specific to Ask mode:
palabre config --ask-summary-agent opencode
To also choose the model:
palabre codex-ollama "Critique this plan" --summary-agent ollama-local --summary-model <installed-ollama-model>
Disable the summary
palabre codex-claude "Critique this plan" --no-summary
Readability
In the terminal, the standard rendering formats the summary headings to make them easier to scan.
In the .debate.md or .ask.md export, the final summary is separated from the transcript or responses by a horizontal rule and begins with a table indicating the agent, role, and production date.
Persistent settings and failure
palabre config --summary-agent <agent> selects the default summary agent; palabre config --summary-agent none removes that choice and restores fallback, without disabling summaries. --no-summary disables the summary for this run. --summary-model passes a raw value to the CLI, which validates the model; Ollama can check installation.
If the summary fails, PALABRE attempts to preserve the transcript in a partial export. With --checkpoint, palabre resume <session-id> resumes only the missing summary, without replaying answers.